Khmer is a low-resource language characterized by a complex script, presenting significant challenges for optical character recognition (OCR). While document printed text recognition has advanced because of available datasets, performance on other modalities, such as handwritten and scene text, remains limited by data scarcity. Training modality-specific models for each modality does not allow cross-modality transfer learning, from which modalities with limited data could otherwise benefit. Moreover, deploying many modality-specific models results in significant memory overhead and requires error-prone routing each input image to the appropriate model. On the other hand, simply training on a combined dataset with a non-uniform data distribution across different modalities often leads to degraded performance on underrepresented modalities. To address these, we propose a universal Khmer text recognition (UKTR) framework capable of handling diverse text modalities. Central to our method is a novel modality-aware adaptive feature selection (MAFS) technique designed to adapt visual features according to a particular input image modality and enhance recognition robustness across modalities. Extensive experiments demonstrate that our model achieves state-of-the-art (SoTA) performance. Furthermore, we introduce the first comprehensive benchmark for universal Khmer text recognition, which we release to the community to facilitate future research. Our datasets and models can be accessible via this gated repository\footnote{in review}.
Figures & tables
Figure 1: Sample images from our new datasets.
Figure 2: Sample Khmer text layout (with permission [ 8 ] .) Blue: consonant subscript. Green: base consonant. Orange: dependent vowel. Purple: diacritic. Best viewed in color.
Figure 3: The proposed UKTR framework.
Name
Params.
Flops
ResNet backbone (visual features)
13.0M
3.2G
Transformers encoder (sequential features)
9.5M
2.2G
Total
22.5M
5.4G
Table 1: The visual encoder’s complexity and model size. Params: model parameters. Flops: floating operations on an input of 32×116 pixels.
Figure 4: The designs of the ROUTER and ADAPTER networks.
Name
Params.
Flops
Transformers decoder (AR decoding)
24.08M
0.82G
Table 2: The Transformers decoder’s complexity and model size. Flops: floating operations on an input of 32×116 pixels.
Figure 5: Sample cropped images from the existing datasets. (a) Buoy et al. [ 1 ] . (b) KHOB. (c) HierText. (d) KhmerST.
Figure 6: Sample images from the WildKhmerST dataset and our GKST dataset.
Figure 7: Sample images from our KHT dataset.
Figure 8: Sample cropped images from our new datasets.
Dataset
Train
Eval
Training (D)
Adapting (S&H)
Eval
Doc
Scene
Handwritten
Buoy et al. [ 1 ]
2.8M
–
✓
–
–
✓
✓
–
KHOB [ 23 ]
–
1,318
–
✓
✓
✓
–
–
SynthText [ 22 ]
70k
–
✓
–
–
✓
–
–
HierText [ 20 ]
518,726
–
✓
–
–
–
✓
✓
KhmerST [ 3 ]
–
3,022
–
✓
✓
–
✓
–
WildKhmerST [ 5 ]
29,601
–
–
✓
–
–
✓
–
Table 3: Summary of datasets, sizes, usage, and modalities.
Dec.
Tok.
KHOB(D)
KhmerST(S)
KH(D,H)
GKST(S)
KHT(H)
Tesseract-OCR [ 11 ]
CTC
Char.
9.19
40.96
–
–
–
Surya-OCR [ 10 ]
Tr.
17.69
43.21
–
–
–
Buoy et al. [ 1 ]
Tr.
Char.
3.03
–
–
–
–
Buoy et al. [ 9 ]
CTC
KCC
2.33
–
–
–
–
Nom et al. [ 3 ]
Tr.
–
17.00
–
–
–
Soy et al. [ 21 ]
Tr.
–
–
17.00
–
–
Table 4: Character error rates (CER) on the evaluation datasets of different modalities (D,S,&H). * : model-specific tokenizer. Char.: character. KCC: Khmer character cluster. Bold : best. Italic : second best.
Decoder
Tokenizer
KHOB
KhmerST
KH
GKST
KHT
UKTR w/o MAFS (D + S&H)
CTC
KCC
3.85
4.30
8.17
6.15
11.89
UKTR w/o MAFS (D + S&H)
Tr.
KCC
3.93
5.48
5.84
5.39
7.66
UKTR w/ MAFS (D + S&H)
CTC
KCC
2.46
3.02
5.89
4.41
9.52
UKTR w/ MAFS (D + S&H)
Tr.
KCC
2.37
2.19
4.11
3.34
6.10
Table 5: Character error rates (CER) w/ and w/o the MAFS module. * : model-specific tokenizer. Char.: character. KCC: Khmer character cluster. Bold : best. Italic : second best.
Decoder
Tokenizer
KHOB
KhmerST
KH
GKST
KHT
UKTR ( n=3 , D + S&H)
CTC
KCC
2.43
3.03
7.12
4.57
9.87
UKTR ( n=3 , D + S&H)
Tr.
KCC
2.29
2.12
5.17
2.99
5.82
UKTR ( n=5 , D + S&H)
CTC
KCC
2.46
3.02
5.89
4.41
9.52
UKTR ( n=5 , D + S&H)
Tr.
KCC
2.37
2.19
4.11
3.34
6.10
UKTR ( n=8 , D + S&H)
CTC
KCC
2.45
3.09
6.50
4.79
9.71
UKTR ( n=8 , D + S&H)
Tr.
KCC
2.07
2.31
4.67
3.78
6.44
Table 6: Character error rates (CER) for n=3 , n=5 , and n=8 . * : model-specific tokenizer. Char.: character. KCC: Khmer character cluster. Bold : best. Italic : second best.
Figure 9: Sample images, ground-truth texts, and model predictions (CTC vs. Transformers).
Text recognition, or extracting electronic text from document images, has been indispensable for knowledge retrieval tasks, such as retrieval-augmented generation (RAG). For Khmer, extracted text is subject to an extra word segmentation step, as Khmer does not use any visible word delimiters to denote word boundaries. Thus, a recognition-then-segmentation pipeline for Khmer requires two separate sequential models; this is not only error-prone but also adds significant latency for large-scale document processing. This paper proposes a novel joint Khmer text recognition and word segmentation framework in a unified model. The proposed model, using a connectionist-temporal-classification (CTC) decoder for fast, parallel decoding, can be instructed to recognize Khmer text with (b=1) and without (b=0) word segmentation. Experimental results on different benchmark datasets of different document modalities (document, scene, and handwritten images) show that the proposed model can not only recognize characters in document images but also locate word boundaries, removing the need for an extra word segmentation step in a conventional sequential pipeline.
Marry Kong, Rina Buoy, Sovisal Chenda +3
Techo Startup Center, Ministry of Economy and Finance, Phnom Penh, Cambodia · Osaka Metropolitan University, Osaka, Japan
While document layout analysis for Latin scripts has advanced significantly, driven by the advent of large multimodal models (LMMs), progress for the Khmer language remains constrained because of the scarcity of annotated training data. This gap is particularly acute for scene documents, where perspective distortions and complex backgrounds challenge traditional methods. Given the structural complexities of Khmer script, such as diacritics and multi-layer character stacking, existing Latin-based layout analysis models fail to accurately delineate semantic layout units, particularly for dense text regions (e.g., list items). In this paper, we present the first comprehensive study on Khmer scene document layout detection. We contribute a novel framework comprising three key elements: (1) a robust training and benchmarking dataset specifically for Khmer scene layouts; (2) an open-source document augmentation tool capable of synthesizing realistic scene documents to scale training data; and (3) layout detection baselines utilizing YOLO-based architectures with oriented bounding boxes (OBB) to handle geometric distortions. To foster further research in the Khmer document analysis and recognition (DAR) community, we release our models, code, and datasets in this gated repository (in review).
Marry Kong, Rina Buoy, Sovisal Chenda +3
Techo Startup Center, Ministry of Economy and Finance, Cambodia · Osaka Metropolitan University, Japan
Optical character recognition (OCR) has advanced rapidly with deep learning and multimodal models, yet most methods focus on well-resourced scripts such as Latin and Chinese. Ethnic minority languages remain underexplored due to complex writing systems, scarce annotations, and diverse historical and modern forms, making generalization in low-resource or zero-shot settings challenging. To address these challenges, we present OmniOCR, a universal framework for ethnic minority scripts. OmniOCR introduces Dynamic Low-Rank Adaptation (Dynamic LoRA) to allocate model capacity across layers and scripts, enabling effective adaptation while preserving knowledge.A sparsity regularization prunes redundant updates, ensuring compact and efficient adaptation without extra inference cost. Evaluations on TibetanMNIST, Shui, ancient Yi, and Dongba show that OmniOCR outperforms zero-shot foundation models and standard post training, achieving state-of-the-art accuracy with superior parameter efficiency, and compared with the state-of-the-art baseline models, it improves accuracy by 39%-66% on these four datasets. Code: https://github.com/AIGeeksGroup/OmniOCR.